[Explanation] How to Completely Create Character Voices and Outfits Using Only AI
Is it possible to create all of a character's voices and outfits solely on your own PC? To get straight to the point, it is fully achievable by combining current local AI technologies.
- Safe character production using image generation AI
- Methods for generating 'fictional voices' using voice AI
- Gacha and dress-up design to enhance gameplay
1. Creating Diverse Outfits with Image Generation AI
In this project, we prepared 6 characters, each with 10 types of outfits (a total of 60 illustrations). These images were created using 'Forge', an image generation AI that runs on my own PC.
A crucial point in production is the control of instructions (prompts). According to official information, it is said that by pre-entering specifications to exclude things like revealing clothes or childish appearances, one can achieve safe and intended character designs.
By giving specific constraints to the AI in this way, it becomes possible to mass-produce content without breaking the world-view that fits your personal preferences.
- Tools used: Forge (local image generation AI)
- Production volume: 6 characters x 10 types of outfits
- Key point: Entering specifications to exclude specific elements
2. Tips for 'Voice Generation' Without Using Any Real Human Voices
Regarding the voices that breathe life into the characters, we do not use any real human voices here at all. We used the voice AI 'Qwen3-TTS' and built everything through generation from text.
As for the specific procedure, we first transcribe the characteristics of 'what kind of voice it is' into text based on the character's profile. Based on that description, we have the AI generate a voice, and then follow up by having it read lines in that voice.
A key technique here is 'reading adjustment'. Since the AI may not be able to pronounce Mahjong terms (Reach, Han, etc.) correctly, we deliberately replace them with 'kana' when inputting them to reproduce accurate pronunciation. A point of the design is that the volume of the voices can also be adjusted separately from other sound effects.
- Tools used: Qwen3-TTS
- Generation method: Directly generating voices from character settings (text)
- Reading ingenuity: Replacing Mahjong terms with 'kana' for input
3. Gacha and Dress-up Systems Supporting User Experience
It is important not only to create materials with AI but also the game design of how to let people 'play' with them. This time, we introduced a system that can be enjoyed without spending money.
The mechanism for obtaining coins is such that they accumulate in the same amount as the experience points earned during play. Regarding the gacha specifications, it is designed so that at least one SR or higher rarity is guaranteed in a 10-roll gacha, and an SSR is guaranteed after 60 pulls. Furthermore, we have arranged for duplicate items to be returned as coins, ensuring that users can enjoy collecting without stress.
The objects to collect are not just characters (avatars), but also suit patterns, reach sticks, table mats, etc., with a total of 27 types prepared. Since the selected character is reflected in name tags and production screens, it allows for play with a sense of attachment.
- Gacha specifications: 1 SR or higher guaranteed per 10-roll / SSR guaranteed at 60 pulls
- Collection elements: Total of 27 types including characters, suit patterns, reach sticks, mats, etc.
- Key point: Design where coins are accumulated via experience points without spending money
Summary: Creating 'Your Own World' with Local AI
In this experiment, we confirmed that by combining image generation AI and voice AI, it is possible to build original character content while minimizing concerns regarding copyright and costs.
The strength of running on your own PC (local environment) lies in the ability to freely repeat these trials and errors. In the next 'Part 4', we will explain how to establish Mahjong as a 'game' using these materials, from the aspects of production and growth systems.
- Both images and voices are completed within your own PC environment
- Proper instructions (constraints) to the AI determine the quality
- Next time: Explanation of 'production and growth systems' scheduled
From the Editorial Department
Through this experiment, we reaffirmed that AI is not just a convenient tool, but a powerful partner that expands individual creativity. What kind of characters or world-views would you like to create using AI? Please let us know in the comments.
- From the shachiku's laboratory
- People interested in AI but don't know where to start
- People who want to try running AI on their own PC
- People who want to streamline content production using AI
- Local AI
- Artificial intelligence that runs directly on a device such as your own PC, rather than on a cloud service.
- Forge
- One of the software packages for efficiently running image generation AI.
- Qwen3-TTS
- A type of model that outputs audio corresponding to the input text (Text-to-Speech).
- Prompt
- Instructional text provided to an AI.
FAQ
Are real human voices used?
No, according to official information, no real human voices are used at all; instead, voices generated by AI from text based on character settings are used.
Does it cost money for the gacha?
No, it does not cost money. It is designed so that coins accumulate in the same amount as the experience points earned while playing.
In this article, we explain methods for completely creating character voices and outfits by utilizing local AI. We introduce specific behind-the-scenes production details such as mass-producing safe illustrations with image generation AI (Forge), creating 'fictional voices' using voice AI (Qwen3-TTS), and gacha design to establish them as a game.